Always listening
Voice transcribed in under 500ms. No app. No buttons.
Listens. Sees. Remembers.
Visual understanding
Point at anything. Aura sees and stores what you're looking at.
Never forgets
Every conversation is indexed. Ask in plain language, get the memory back.
Self-host your backend. No cloud account required. Your data stays yours. How privacy works →
The third device.
Not a computer. Not a phone. Something worn.
Not a computer
Not a phone
Something worn
Specifications
Technical specs.
Standard off-the-shelf parts. No markup. MIT licensed.
- 0.5s
- Transcription latency
- Voice-to-text via Groq LPU inference. Fast enough for natural conversation.
- ~$50
- Bill of materials
- Standard off-the-shelf parts. No markup. MIT licensed.
- 80g
- Total weight
- 50 × 68 × 18 mm. 900 mAh Li-Po. ~2 hours active, ~1 hour charge.
Build yours today.
Fork the repo, order the parts, and follow the build guide. No soldering experience required for the base build.
Questions?
- All processing happens on your self-hosted backend. No cloud account required.
- ~4 hours active use, 45 minutes to full charge.
- No soldering required for the base build. The case snaps together.
- Yes. Configure Deepgram, Groq, OpenAI, or local Whisper in the backend .env.
- Yes. MIT licensed. Case STLs, firmware, and backend on GitHub.
- Any phone with Bluetooth LE. The Omi app runs on iOS and Android.